Answers are grounded in this paper and cite their sources.
Pesticide residue status is a key compliance attribute of fresh fruits. For certified agricultural products, label claims should be consistent with residue levels and provide assurance of safety to consumers. On-site monitoring is not a certification tool, but it can support rapid screening and compliance verification. In this study, a portable visible–near-infrared (Vis–NIR) spectroscopy device (instrument range 190–1100 nm; analyzed range 300–950 nm) was developed specifically for rapid, non-destructive authentication of pesticide residue status in longan. The device integrates acquisition, control, power, and display modules. Reference quantitative analysis using QuEChERS and LC-DAD anchored the classification labels to actual residue concentrations and the maximum residue limits (MRLs) specified in GB 27632026, enabling the device to flag potentially non-compliant samples. After preprocessing with Savitzky–Golay smoothing and first derivative (SG-D1), a convolutional neural network (CNN)-based classification model was established for two authentication tasks: pesticide type and residue level. SG-D1 improved classification accuracy by 1.10% and 5.42% compared with raw spectra. Feature wavelengths were selected using the successive projections algorithm (SPA) and competitive adaptive reweighted sampling (CARS). Combined models of CNN, support vector machine (SVM), backpropagation neural network (BPNN), and random forest (RF) were compared. Among feature-selection models, CARS+CNN achieved the best balance between accuracy and computational efficiency. The optimal model was embedded into the device with real-time software for one-click detection. Field tests on 80 longan samples showed accuracies of 98.75% for pesticide type, 92.5% for residue level, and 96.25% for MRL compliance classification, with a single-sample detection time of 6.8 s. This portable device provides a practical tool for the rapid on-site screening of pesticide residue compliance in food.
No takes yet. Share an insight, caveat, or question.
Li et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: